Short-term oriented oil price models and the financial crisis

Transcription

1 Short-term oriented oil price models and the financial crisis The need for oil price models with a short-term forecasting horizon has increased considerably in recent years, not least with the emergence of Exchange Traded Funds (ETFs) in the commodity sector, i.e. exchangetraded commodity investment funds with no fixed maturity. In this paper, we present three different forecasting models on a weekly basis. They permit us to make an oil price prediction with a time horizon of up to three months. The first two variants are so-called Vector Auto-Regressive (VAR) models and incorporate fundamental factors such as the net long positions or oil stockpiles. The third variant is a pure Futures model. It becomes apparent that the two fundamental models generate superior results up to mid/end-2007 and from mid-2009 on. In contrast, the Futures model is clearly superior in predicting the oil price during the financial market crisis from the end-2007 to mid Growing need for short-term models Because of the strong oil price increases primarily at the beginning of the century, the oil market has attracted increasing attention from various quarters, and specifically from banks. This has impacted the range of products offered (e.g. ETFs on commodities), but also the need to develop quantitative models to forecast the oil price. The latter trend was amplified further by the price slumps with the onset of the financial market crisis in 2007 and the renewed rise in the oil price from the turn of 2008/09. Weekly and monthly models are prime candidates for those interested primarily in short-term oil price forecasts (up to at most one quarter). The academic literature differentiates here between pure financial models, which use only information from spot and futures prices, and structural models (fundamental models), which factor in the special characteristics of the oil market (specifically, the supply and demand situation and its determinants). Longo et al. (2007) provides a good overview here, including further differentiations above all of the empirical-econometric approach. In the following, we present three different weekly models from both categories for the oil price (West Texas Intermediate, WTI). Our primary interest is the performance of the models in quiet versus turbulent phases (as during the recent financial market crisis). Subsequently, we describe the models, the econometric methodology and the data used. We then discuss in detail the results and the performance of the models since mid Econometric methodology In principle, there are two alternatives available to explain and predict the oil price. The first is a univariate approach. This has the advantage that it is easy to model and that the results are, as a rule, easier to interpret and more plausible for the user. If, however, the primary focus is on the forecast, univariate approaches have the drawback that the exogenous variables also have to be predicted, thereby creating an additional area of uncertainty. Authors Jörg Clostermann (University of Applied Sciences, Ingolstadt) clostermann@fh-ingolstadt.de Nikolaus Keis (UniCredit Bank), nikolaus.keis@unicreditgroup.de Franz Seitz (University of Applied Sciences, Weiden, WSB Poznan), f.seitz@haw-aw.de Bloomberg UCGR, UCFR Internet UniCredit Research page 1 See last pages for disclaimer.

2 Vector Auto-Regressive Models For that reason, the scientific norm now is the use of state-of-the-art, Vector Auto-Regressive (VAR) models, in which all variables used are endogenized (with respect to oil prices see, for example, Akram, 2009; Kaufmann et al., 2004; Miller and Ratti, 2009). Because of the concrete model structure, each variable from the model context is itself forecasted within the model for arbitrary forecasting horizons. It was possible to demonstrate that an unrestricted VAR is a good approximation for every data-generating process, provided sufficient lags are factored into the variables (Canova, 1995). A VAR takes the following form: (1) t =Φ+ 1 t p t p + t X AX A X ε where X t represents the vector of the endogenous variables, Φ represents the matrix of the deterministic terms, specifically the constant and a linear deterministic trend, A 1 to A p represent the symmetric coefficient matrices, p represents the selected lag length and ε t the vector of the residuals. If the variables used are not stationary but cointegrated, it is possible to recast the VAR model into a so-called Vector Error Correction (VEC) model (for details, see Johansen, 1995). To this end (1) is modified to (2) p 1 Δ X =Φ+Π X + Γ Δ X + ε t t 1 i t i t i= 1 p mit Π= A I, Γ = A i i j i= 1 j= i+ 1 p In (2), X t is the vector of the k non-stationary I(1) variables. If the matrix has Π reduced ranking (r-k), there are according to Granger's representation theorem (kxr) matrices α and ß with ranking r, with the result that Π = αβ' and ß'X t I(0); r represents here the number of cointegration relationships and each column of ß includes a cointegration vector. The α- coefficients are the so-called adjustment coefficients or error correction terms of the VEC model. Here, the number of cointegration relationships is normally determined by the trace and maximum eigenvalue statistics. Regime shift in 1999 Data and forecast approaches The objective of the model is to explain and forecast the oil price in US dollars per barrel. As is customary, we use the respective "nearest future" contract. Its development is shown in the following chart on a weekly basis (in each case the Friday reading) since the beginning of To eliminate erratic fluctuations because of the cut-off date, we form moving 4-week averages. The rather moderate development up to the beginning of the 21st century is clearly evident; the price increases that subsequently ensued emerge clearly from the end of 2006 to the beginning of 2008 and resulted in a oil price of over USD 140 per barrel. Subsequently, the oil price fell to below USD 40 by the beginning of 2009, only to rise to levels of up to USD 80 again by the end of The graphic representation of the oil price (cf. chart next page) suggests that we have been in a new regime since Because of these developments, we decided to start the investigation period only in Overall, therefore, we have 565 observations at our disposal. We use a weekly model so as to be able to analyze forecasting horizons of one week up to three months. Fundamental models the role of oil inventories... Because of the selection of a weekly oil price, the explanatory variables should also be available at least in this frequency. As a fundamental factor in this context, the crude oil stockpiles are traditionally analyzed as "summary statistic" for the supply/demand situation (see, for example, Chevillon and Rifflart, 2009; Kaufmann et al., 2004; Ye at al., 2005; Zamani, 2004). In the following, we use the industrial stockpiles of the US, since these are the only consistent data available on a weekly basis. They constitute the bulk of the OECD UniCredit Research page 2 See last pages for disclaimer.

3 stockpiles. The theoretical correlation between the stockpiles and the oil price is, however, not conclusive. A negative correlation is based on the view that rising stockpiles indicate a supply surplus, which has a dampening effect on the oil price. In contrast, strategic considerations argue for a positive correlation: The expectation of rising oil prices could, for example, prompt the industrialized countries to increase their stockpiles. Security considerations could also play a role here.... and speculative net long positions In addition to the stockpiles, we use further time series that are available on a weekly basis. First, we include the net long position of non-commercial traders (see, for example, also Merino and Ortiz, 2005). Rising net long positions suggest that the oil price will increase. However, interdependencies between the net long position and the stockpiles must also be taken into account. Under certain circumstances, changed stockpiles impact investor behavior, and vice versa. This is, for example, suggested by the positive correlation between the net long position and the stockpiles (cf. chart below). Depending on the period observed and the factored lag, the correlation coefficient between both time series was in some cases up to 50%. Natural gas as a substitute As a further variable, we factor in the Henry-Hub natural gas price in US dollars per MMBTU ("million British thermal units") and gasoline consumption in thousands of barrels per day. Since natural gas functions as a substitute for crude oil, there should be a positive correlation between the natural gas price and the oil price. Equally, rising demand for gasoline should drive up the price of crude oil. Both variables are illustrated in the chart next page. It is apparent that gasoline consumption has shifted only marginally higher since 1999, albeit with clear fluctuations from week to week. These volatilities do not appear to have changed. The natural gas price demonstrates clearer persistence, albeit with pronounced short-term fluctuations. OIL PRICE DEVELOPMENTS NET LONG POSITION AND US INDUSTRIAL STOCKPILES Nearest future Oil Price (WTI, USD/b) 4W moving average 350, , ,000 Net-long position (contracts) US crude oil inventories (commercial, 1000 barrels, RS) , , ,000 50, , /93 01/95 01/97 01/99 01/01 01/03 01/05 01/07 01/09-100,000 01/99 07/00 01/02 07/03 01/05 07/06 01/08 07/ Source: Bloomberg, Thomson Datastream, UniCredit Research Futures model Alongside these VAR models, we have estimated a pure financial market model, which exploits only the relationships between oil futures of various maturities and the spot price. Coppola (2008), for example, found evidence supporting a cointegration relationship between the oil spot price and the futures prices and a high explanatory power of futures prices for spot prices (see also Abosedra, 2005; Dées et al., 2008; Huang et al., 2009). If price innovations become evident initially in spot prices, market fundamental data are probably decisive for the crude oil price development. If, in contrast, the futures prices react first, speculation likely assumes an important role (Kaufmann and Ulmann, 2009). Alongside the spot price and the UniCredit Research page 3 See last pages for disclaimer.

4 "nearest future" price, we also include the 2M and 3M future. They are shown in the following chart. Arbitrage processes produce the extreme parallel movement of the three time series. Furthermore, statistical tests show that they are profoundly non-stationary. NATURAL GAS PRICE AND GASOLINE CONSUMPTION OIL PRICE FUTURES US gasoline consumption (1000 barrels per day) Natural gas price (Henry Hub, USD/MMBTU*, RS) /99 01/01 01/03 01/05 01/07 01/ M Future (in % of the nearest Future) 3M Future (in % of the nearest Future) /4/1999 1/4/2001 1/4/2003 1/4/2005 1/4/2007 1/4/2009 Source: Bloomberg, Thomson Datastream, UniCredit Research Expectations theory as a theoretical foundation The theoretical basis of the correlations is the expectations theory, which is based on their time series characteristic. If the various oil prices follow a stochastic trend, they must all follow the same trend irrespective of the maturity, i.e. the price differences between (arbitrarily selected) price pairs are stationary. In our case, the expectations theory states specifically that in the arbitrage equilibrium the "long-term" price for example measured by the 3M future must be identical to the price expected from revolving short-term transactions for example, three consecutive transactions with a one-week maturity. With P t as 3M future, p t as 1M future, this then produces ("e" characterizes expectations variables) (3) P 1 e e t = t + t+ 1+ t+ 2 3 (p p p ) If p t is subtracted on both sides, one obtains (4) P p = Δp + + Δp +, 2 e 1 e t t 3 t 1 3 t 2 with Δp t+1 = p t+1 p t as one-period price change. The spread P t p t is, therefore, a weighted mean of the changes expected in the 1M future over the next two months, whereby the price changes expected further in the future assume a smaller weight. This is illustrated by the expectations content, which according to the expectations theory lurks in the difference between a long-term and short-term "price". Above and beyond that, it is apparent from (4) that the long-term and the short-term prices follow a common trend, i.e. the spread between the 3M future and the 1M future is stationary. On the right side of (4) stand only (expected) changes of the short-term future. Since the futures follow a stochastic trend, the price difference and therefore also the right side of (4) are consequently stationary. In fact, multivariate cointegration tests show this to be the case. In our sample, the spreads we observed become greater as the maturity difference becomes longer. That also means that spot prices are, in general, lower than the future prices, which is commonly referred to as the contango effect. The marginal "convenience yields" are then probably relatively high. The socalled backwardation, spot prices above the futures prices, a standard result in the literature (e.g. French, 2005) does not, therefore, hold generally for our data set. Huang et al. (2009) also find this result for the period from September 11, 2001 to April 30, 2007, which covered the largest part of our sample. Kaufmann et al. (2008) conclude that the strong rise in the UniCredit Research page 4 See last pages for disclaimer.

5 price of oil from 2005 to 2007 went hand in hand with a change in the futures markets from backwardation to contango. General-to-specific approach Out of sample forecasts With the exception of the net long position, all variables are logarithmed. To soften the impact of individual outliers on the results and to better correlate the time series, we calculate moving 4-week averages from all variables. The lag length in the VAR models is designed to maintain forecasting quality despite frugal modeling. The respective final model includes only those variables that are required in each case on forecasting considerations. That also means that even if an additional variable or an additional lag is significant but does not bring an improvement in the forecasting, it is not included in the approach. Here, we proceed according to the so-called "general-to-specific" methodology. The forecasting period extends at most from the beginning of 1999 to October 26, Our primary interest here is how the quality of the models changed during the financial market crisis since mid/end Forecasting For out-of-sample forecasts, the respective forecasting period is shortened by one year, and the "missing" year is used as the forecasting period. The estimates and forecasts are conducted recursively, where the period used is in each case lengthened by one week. The forecasting horizons extend from one week to three months (12 weeks). The forecasting quality of our three models (VAR, VEC, Futures) is assessed using several forecasting quality yardsticks. The benchmarks used for each sample are an optimized Random Walk with/without drift and an ARIMA model. For each of these models, the oil price was estimated from the beginning of 1999 to point t and forecasted up to h weeks into the future based on the respective current data position (=wti t+h,t )). The forecasting error (=e t+h,t ) is then the difference between the actual value at point t+h (=wti t+h ) and its forecasted value: e + = wti wti + (5) t h,t t+ h t h, t The comparison of the forecasted values with the actual values produces the Root Mean Squared Error (RMSE). It is defined as The second criterion used is the so-called Direction Hit Ratio (DHR). To this end, an indicator variable J is defined with the following characteristics (6) RMSE 1 T = 2 h et+ h, t T t = 1 ( t+ h t) ( t+ h, t t) ( t+ h t) ( t+ h, t t) if sign wti wti = sign wti wti J = 1 if sign wti wti sign wti wti J = 0 Consequently, the DHR equals VZT h T 1 = Jt 100% T t = 1 The higher the DHR, the greater the frequency that the direction of the oil price change is predicted correctly. For example, a direction hit ratio of 70% indicates that in 70% of cases the model correctly predicted the direction. Both forecasting quality yardsticks RMSE and DHR are discussed in Cheung et al. (2005). Furthermore, we assess a still rather unusual measure which we call the "Mean Weighted Hit Ratio" (MWHR). It in turn refers back to the indicator function (6); only false hits are now allocated the value "-1". MWHR is defined as follows ((1, 1) ) MWHR = mean Δ wti UniCredit Research page 5 See last pages for disclaimer.

6 The higher the values, the better the forecasting quality. MWHR weights the direction hit ratio with the absolute extent of the change of the oil price at the corresponding point in time. Extremely positive is factored in if major changes are predicted correctly, but extremely negative is factored in if major changes are not predicted correctly. Model results I: The behavior up to the end of 2007 The models were optimized for the period up to the financial market crisis at mid/end Alongside the oil price, the only variables included (in the levels) in the two fundamental models are the natural gas price and the net long position. The net long position is included as an exogenous, not-modeled variable, since compared to the other variables it is already stationary in the levels. For the forecast, it is modeled as a univariate process. The VEC model factors in a cointegration relationship between the oil and the natural gas price. The Futures VEC model includes pursuant to the theoretical considerations discussed in the preceding section two cointegration relationships between the three oil prices. The two VEC models include three lags in the levels. It was not necessary to model seasonal effects in any of the models, since no pronounced and systemic seasonal patterns were observable (see here, for example, Zamani, 2004). Fundamental models superior during quiet market phases Tables 1-3 show the results for the three forecasting quality yardsticks (RMSE, DHR, MWHR) and the four models (VAR, VEC, Futures, Random Walk (RW)) for the 12 forecast horizons. It is apparent that the Random Walk is beaten by all three models, irrespective of the forecasting measure. With respect to futures prices, this result is also confirmed by Abosedra (2005) as well as Murat and Tokat (2009). In terms of the absolute hit ratio, measured by the RMSE, the VAR and the VEC model perform better on forecasting horizons of up to two months and better than the Futures model. The quality of the results generally declines with the forecast horizon. In the case of the DHR, the VEC fundamental model performs best irrespective of the forecast horizon. Specifically for maturities over two months, this model correctly predicts up to 80% of the oil price changes. The VAR and the Futures model produce clearly poorer results in this respect. That is also confirmed by Table 3, which shows the MWHR. The comparatively higher forecasting quality of the fundamental error correction model argues for the validity of the assumed cointegration relationship. Overall, this analysis points to the conclusion that up to the end of 2007 the two fundamental models, specifically the VEC model, produce pretty good results, and are clearly superior to the Futures model but also a Random Walk. The correlations could, however, have changed with the onset of the structural break triggered by the financial market crisis. For that reason, the focus in the following is on the performance of the various models from 2008 to the end of TABLE 1: ROOT MEAN SQUARED ERROR (RMSE) Forecast horizon VAR VEC RW Futures 1 week weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks UniCredit Research page 6 See last pages for disclaimer.

7 TABLE 2: DIRECTION HIT RATIO (DHR) Forecast horizon VAR VEC RW Futures 1 week weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks TABLE 3: MEAN-WEIGHTED HIT RATIO (MWHR) Forecast horizon VAR VEC RW Futures 1 week weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks weeks Source: UniCredit Research Model results II: Behavior during the financial market crisis Futures model superior during turbulent market phases The following charts (next page) compare the forecast produced by our three models for a 3- month period with the actual development of the oil price for selected periods from mid The main focus of interest is the turning points (mid-2008, beginning of 2009). The process is once again recursive, and the period observed is extended successively to the current edge (end of October 2009). It is clear that the Futures model clearly dominates up to mid-2009, but that from the second half of 2009 the two fundamental models again come out on top. Even though the sharp pullback by the oil price in 2008 (sample up to week 26, 2008) is shown only weakly by the futures, the VAR and VEC model would, however, still have signaled rising prices. The renewed rise from February 2009 is, however, signaled early on by the Futures model (sample up to week 52, 2008). While this ranking persists until mid-2009 (sample up to week 26, 2009), the superiority of the two fundamental models is evident again after the gradual calming of the situation on financial markets (sample up to week 30, 2009). If the period observed is extended up to the end of October (sample: total period observed), the Futures model would predict only marginally rising oil prices until the end of January 2010, while the two other models suggest prices rising to over USD 90. UniCredit Research page 7 See last pages for disclaimer.

8 MODEL FORECASTS DURING THE FINANCIAL CRISIS , week , week WTI VEC VAR FUTURE 90 6/16/08 7/1/08 7/16/08 7/31/08 8/15/08 8/30/08 9/14/08 40 WTI VEC VAR FUTURE 35 12/15/08 1/5/09 1/26/09 2/16/09 3/9/ , week , week WTI VEC VAR FUTURE 64 6/15/09 7/6/09 7/27/09 8/17/09 9/7/09 64 WTI VEC VAR FUTURE 62 7/13/2009 8/03/2009 8/24/2009 9/14/ /05/ total sample period VEC VAR FUTURE 10/12/09 11/2/09 11/23/09 12/14/09 1/4/10 Note: Above the corresponding figures is the end of the respective period observed. Shown are the 4-week moving averages. Source: Bloomberg, Thomson Datastream, UniCredit Research Overall, these observations permit the conclusion that during strong upheavals on markets, characterized by extreme shocks, preference should be given to the Futures model over the two other models. In quiet phases, in contrast, the fundamental VAR and VEC models are more reliable. The models also appear to be still valid; the financial market crisis probably did not trigger any basic changes or instabilities. Only the short-term performance was affected. UniCredit Research page 8 See last pages for disclaimer.

9 Don't put all your eggs in one basket Summary and outlook In this analysis, we have presented three different forecasting models for the oil price. Model selection was based on a purely forecast-oriented approach. Accordingly, VAR models formed the methodological-econometric basis. On the one hand, we factored in fundamental determinants on a weekly basis, such as the net long position and the natural gas price. This was, on the other hand, compared with a pure financial market model based on futures prices. It was evident that a VAR or VEC model with fundamental variables produces good forecasting results in quiet market phases and is clearly superior to a Random Walk model and also to the Futures model. In turbulent market phases, such as triggered by the financial market crisis, investors should however tend to rely on the Futures model. It is, therefore, advantageous overall to monitor both variants and focus on a specific model depending on the market situation. It was astonishing that the stockpiles had no significant influence and were not included in the ultimately preferred fundamental models. This could have to do with the weekly data used, but also with the fact that only US industrial stockpiles are available on a weekly basis and that we also included the net long position. In the literature, a significant influence of the stockpiles is found if these assumptions are sacrificed (see, for example, the overview in Longo et al., 2007, chapter 2.2). Some authors do, however, note that it is not the stockpiles per se but their position relative to a normal level that is the decisive variable (Ye et al., 2005; Zamani, 2004). The influence of the stockpiles could also depend on whether the market is in a phase of rising or falling prices (Ye et al, 2005). Geman and Ohana (2009) also find that the information in futures prices is a good proxy for stockpiles. The stockpiles would, therefore, be implicitly included in the Futures model. It will require further research to determine whether the correlations discovered here are also to be found in models with monthly or quarterly frequency. Above and beyond that, it is interesting to investigate to what extent models with low data frequency are more stable in turbulent market phases, and whether they are also suitable as valid approaches for longerterm forecasts. UniCredit Research page 9 See last pages for disclaimer.

10 Literature Abosedra, S. (2005), Futures versus Univariate Forecast of Crude Oil Prices, OPEC Review 29, S Akram, Q.F. (2009), Commodity Prices, Interest Rates and the Dollar, Energy Economics 31, S Canova, F. (1995), VAR: Specification, Estimation, Testing and Forecasting, in: Pesaran, H., Wickens, M. (Hg.), Handbook of Applied Econometrics, Blackwell, London, S Chevillon, G. & C. Rifflart (2009), Physical Market Determinants of the Price of Crude Oil and the Market Premium, Energy Economics 31, S Cheung, Y.-W., M.D. Chinn & A.G. Pascual (2005), Empirical Exchange Rate Models of the Nineties: Are any fit to survive?, Journal of International Money and Finance 24, S Chinn M.D., LeBlanc, M & O. Coibion (2005), The Predictive Content of Energy Futures: An Update on Petroleum, Natural Gas, Heating Oil and Gasoline, NBER Working Paper No , Januar. Coppola, A., (2008), Forecasting Oil Price Movements: Exploiting the information in the futures market, Journal of Futures Market 28, S Dées, S., Gasteuil, A., Kaufmann, R.A. & M. Mann (2008), Assessing the Factors behind Oil Price Changes, ECB Working Paper No. 855, Januar. French, M.W. (2005), Why and When do Spot Prices of Crude Oil Revert to Futures Price Levels? Federal Reserve Board, Finance and Economics Discussion Series , Juli. Geman, H. & S. Ohana (2009), Forward Curves, Scarcity and Price Volatility in Oil and Natural Gas Markets, Energy Economics 31, S Hamilton, J. (2009), Understanding Crude Oil Prices, The Energy Journal 30, S Huang, B.-N., Yang, C.W. & M.J. Hwang (2009), The Dynamics of a Nonlinear Relationship between Crude Oil Spot and Futures Prices: A multivariate threshold regression approach, Energy Economics 31, S Johansen, S. (1995), Likelihood-based Inference in Cointegrated Vector Auto-regressive Models, Oxford University Press, Oxford, New York. Kang, S.H., Kang, S.-M. & S.-M. Yoon (2009), Forecasting Volatility of Crude Oil Markets, Energy Economics 31, S Kaufmann, R.K., Dees, S., Karadeloglou, P. & M. Sanchez (2004), Does OPEC Matter: An econometric analysis of oil prices, The Energy Journal 25, S Kaufmann, R.K, Karadeloglou, P & F. di Mauro (2008), Will Oil Prices Decline over the Long Run, ECB Occasional Paper No 98. Kaufmann, R.K. & B. Ulman (2009), Oil Prices, Speculation, and Fundamentals: Interpreting causal relations among spot and futures prices, Energy Economics 31, S Krichene, N. (2005), A Simultaneous Equations Model for World Crude Oil and Natural Gas Markets, IMF Working Paper 05/32, Februar. Lehman Brothers (2007), A Brief History of the Oil Price, Global Equity Research vom Litzenberger, R.H. & N. Rabinowitz (1995), Backwardation in Oil Futures Markets: Theory and Empirical Evidence, Journal of Finance 50, S Longo, C., Manera, M., Markandya, A. & E. Scarpa (2007), Evaluating the Empirical Performance of Alternative Econometric Models for Oil Price Forecasting, IEM Nota di Lavoro , Januar. Merino, A. & Á. Ortiz (2005), Explaining the so-called "Price Premium" in Oil Markets, OPEC Review 29, S Miller, J.I. & R.A. Ratti (2009), Crude Oil and Stock Markets: Stability, instability, and bubbles, Energy Economics 31, S Murat, A. & E. Tokat (2009), Forecasting Oil Price Movements with Crack Spread Futures, Energy Economics 31, S UniCredit Research page 10 See last pages for disclaimer.

11 Reitz, S., Rülke J.C. &G. Stadtmann (2009), Are Oil Price Forecasters Finally Right? Regressive Expectations Toward More Fundamental Values of the Oil Price, Deutsche Bundesbank Discussion Paper, Series 1: Economic Studies, No 32 Pindyck, R.S.(2001), The Dynamics of Commodity Spot and Futures Markets: A Primer, Energy Journal 22, S Shouyang, W., Lean, Y. & Lai, K.K. (2005), Crude Oil Price Forecasting with TEI@I Methodology, Journal of Systems Science and Complexity 18, S Ye, M., Zyren, J. & J. Shore (2005), A Monthly Crude Oil Spot Price Forecasting Model Using Relative Inventories, International Journal of Forecasting 21, S Zamani, M. (2004), An Econometrics Forecasting Model of Short-term Oil Spot Price, Paper presented at the 6th IAEE European Conference UniCredit Research page 11 See last pages for disclaimer.

12 Disclaimer Our recommendations are based on information obtained from, or are based upon public information sources that we consider to be reliable but for the completeness and accuracy of which we assume no liability. All estimates and opinions included in the report represent the independent judgment of the analysts as of the date of the issue. We reserve the right to modify the views expressed herein at any time without notice. Moreover, we reserve the right not to update this information or to discontinue it altogether without notice. This analysis is for information purposes only and (i) does not constitute or form part of any offer for sale or subscription of or solicitation of any offer to buy or subscribe for any financial, money market or investment instrument or any security, (ii) is neither intended as such an offer for sale or subscription of or solicitation of an offer to buy or subscribe for any financial, money market or investment instrument or any security nor (iii) as an advertisement thereof. The investment possibilities discussed in this report may not be suitable for certain investors depending on their specific investment objectives and time horizon or in the context of their overall financial situation. The investments discussed may fluctuate in price or value. Investors may get back less than they invested. Changes in rates of exchange may have an adverse effect on the value of investments. Furthermore, past performance is not necessarily indicative of future results. In particular, the risks associated with an investment in the financial, money market or investment instrument or security under discussion are not explained in their entirety. This information is given without any warranty on an "as is" basis and should not be regarded as a substitute for obtaining individual advice. Investors must make their own determination of the appropriateness of an investment in any instruments referred to herein based on the merits and risks involved, their own investment strategy and their legal, fiscal and financial position. As this document does not qualify as an investment recommendation or as a direct investment recommendation, neither this document nor any part of it shall form the basis of, or be relied on in connection with or act as an inducement to enter into, any contract or commitment whatsoever. Investors are urged to contact their bank's investment advisor for individual explanations and advice. Neither UniCredit Bank AG, UniCredit Bank AG London Branch, UniCredit CAIB AG, UniCredit Bank AG Milan Branch, UniCredit CAIB Securities UK Ltd., UniCredit Securities, UniCredit Menkul Değerler A.Ş., UniCredit Bulbank, Zagrebačka banka, UniCredit Bank, Bank Pekao, Yapi Kredi, UniCredit Tiriac Bank, ATFBank, nor any of their respective directors, officers or employees nor any other person accepts any liability whatsoever (in negligence or otherwise) for any loss howsoever arising from any use of this document or its contents or otherwise arising in connection therewith. This analysis is being distributed by electronic and ordinary mail to professional investors, who are expected to make their own investment decisions without undue reliance on this publication, and may not be redistributed, reproduced or published in whole or in part for any purpose. Responsibility for the content of this publication lies with: a) UniCredit Bank AG, Am Tucherpark 16, Munich, Germany, (also responsible for the distribution pursuant to 34b WpHG). The company belongs to UCI Group. Regulatory authority: BaFin Bundesanstalt für Finanzdienstleistungsaufsicht, Lurgiallee 12, Frankfurt, Germany. b) UniCredit Bank AG London Branch, Moor House, 120 London Wall, London EC2Y 5ET, United Kingdom. Regulatory authority: BaFin Bundesanstalt für Finanzdienstleistungsaufsicht, Lurgiallee 12, Frankfurt, Germany and subject to limited regulation by the Financial Services Authority (FSA), 25 The North Colonnade, Canary Wharf, London E14 5HS, United Kingdom. Details about the extent of our regulation by the Financial Services Authority are available from us on request. c) UniCredit Bank AG Milan Branch, Via Tommaso Grossi, 10, Milan, Italy, duly authorized by the Bank of Italy to provide investment services. Regulatory authority: Bank of Italy, Via Nazionale 91, Roma, Italy and Bundesanstalt für Finanzdienstleistungsaufsicht, Lurgiallee 12, Frankfurt, Germany. The UniCredit CAIB Group, consisting of d) UniCredit CAIB AG, Julius-Tandler-Platz 3, 1090 Vienna, Austria Regulatory authority: Finanzmarktaufsichtsbehörde (FMA), Praterstrasse 23, 1020 Vienna, Austria e) UniCredit CAIB Securities UK Ltd., Moor House, 120 London Wall, London EC2Y 5ET, United Kingdom Regulatory authority: Financial Services Authority (FSA), 25 The North Colonnade, Canary Wharf, London E14 5HS, United Kingdom f) UniCredit Securities, Boulevard Ring Office Building, 17/1 Chistoprudni Boulevard, Moscow , Russia Regulatory authority: Federal Service on Financial Markets, 9 Leninsky prospekt, Moscow , Russia g) UniCredit Menkul Değerler A.Ş., Büyükdere Cad. No. 195, Büyükdere Plaza Kat. 5, Levent, Istanbul, Turkey Regulatory authority: Sermaye Piyasası Kurulu Capital Markets Board of Turkey, Eskişehir Yolu 8.Km No:156, Ankara, Turkey h) UniCredit Bulbank, Sveta Nedelya Sq. 7, BG-1000 Sofia, Bulgaria Regulatory authority: Financial Supervision Commission, 33 Shar Planina str.,1303 Sofia, Bulgaria i) Zagrebačka banka, Paromlinska 2, HR Zagreb, Croatia Regulatory authority: Croatian Agency for Supervision of Financial Services, Miramarska 24B, Zagreb, Croatia j) UniCredit Bank, Na Príkope 858/20, CZ Prague, Czech Republic Regulatory authority: CNB Czech National Bank, Na Příkopě 28, Praha 1, Czech Republic k) Bank Pekao, ul. Grzybowska 53/57, PL Warsaw, Poland Regulatory authority: Polish Financial Supervision Authority, Plac Powstańców Warszawy 1, Warsaw, Poland l) UniCredit Bank, Prechistenskaya emb. 9, RF Moscow, Russia Regulatory authority: Federal Service on Financial Markets, 9 Leninsky prospekt, Moscow , Russia m) UniCredit Bank, Šancova 1/A, SK Bratislava, Slovakia Regulatory authority: National Bank of Slovakia, Stefanikovo nam. 10/19, Kremnica, Slovakia n) Yapi Kredi, Yapi Kredi Plaza D Blok, Levent, TR Istanbul, Turkey Regulatory authority: Sermaye Piyasası Kurulu Capital Markets Board of Turkey, Eskişehir Yolu 8.Km No:156, Ankara, Turkey o) UniCredit Tiriac Bank, Ghetarilor Street 23-25, RO Bucharest 1,Romania Regulatory authority: CNVM, Romanian National Securities Commission, Foişorului street, no.2, sector 3, Bucharest, Romania p) ATFBank, 100 Furmanov Str., KZ Almaty, Kazakhstan Agency of the Republic of Kazakhstan on the state regulation and supervision of financial market and financial organisations, , Almaty, 67 Aiteke Bi str., Kazakhstan POTENTIAL CONFLICTS OF INTEREST UniCredit Bank AG acts as a Specialist or Primary Dealer in government bonds issued by the Italian, Portuguese and Greek Treasury. Main tasks of the Specialist are to participate with continuity and efficiency to the governments' securities auctions, to contribute to the efficiency of the secondary market through market making activity and quoting requirements and to contribute to the management of public debt and to the debt issuance policy choices, also through advisory and research activities. ANALYST DECLARATION The author s remuneration has not been, and will not be, geared to the recommendations or views expressed in this study, neither directly nor indirectly. ORGANIZATIONAL AND ADMINISTRATIVE ARRANGEMENTS TO AVOID AND PREVENT CONFLICTS OF INTEREST To prevent or remedy conflicts of interest, UniCredit Bank AG, UniCredit Bank AG London Branch, UniCredit CAIB AG, UniCredit Bank AG Milan Branch, UniCredit CAIB Securities UK Ltd., UniCredit Securities, UniCredit Menkul Değerler A.Ş., UniCredit Bulbank, Zagrebačka banka, UniCredit Bank, Bank Pekao, Yapi Kredi, UniCredit Tiriac Bank, ATFBank have established the organizational arrangements required from a legal and supervisory aspect, adherence to which is monitored by its compliance department. Conflicts of interest arising are managed by legal and physical and non-physical barriers (collectively referred to as Chinese Walls ) designed to restrict the flow of information between one area/department of UniCredit Bank AG, UniCredit Bank AG London Branch, UniCredit CAIB AG, UniCredit Bank AG Milan Branch, UniCredit CAIB Securities UK Ltd., UniCredit Securities, UniCredit Menkul Değerler A.Ş., UniCredit Bulbank, Zagrebačka banka, UniCredit Bank, Bank Pekao, Yapi Kredi, UniCredit Tiriac Bank, ATFBank and another. In particular, Investment Banking units, including corporate finance, capital market activities, financial advisory and other capital raising activities, are segregated by physical and non-physical boundaries from Markets Units, as well as the research department. In the case of equities execution by UniCredit Bank AG Milan Branch, other than as a matter of client facilitation or delta hedging of OTC and listed derivative positions, there is no proprietary trading. Disclosure of publicly available conflicts of interest and other material interests is made in the research. Analysts are supervised and managed on a day-to-day basis by line managers who do not have responsibility for Investment Banking activities, including corporate finance activities, or other activities other than the sale of securities to clients. UniCredit Research page 12

13 ADDITIONAL REQUIRED DISCLOSURES UNDER THE LAWS AND REGULATIONS OF JURISDICTIONS INDICATED Notice to Austrian investors This document does not constitute or form part of any offer for sale or subscription of or solicitation of any offer to buy or subscribe for any securities and neither this document nor any part of it shall form the basis of, or be relied on in connection with or act as an inducement to enter into, any contract or commitment whatsoever. This document is confidential and is being supplied to you solely for your information and may not be reproduced, redistributed or passed on to any other person or published, in whole or part, for any purpose. Notice to Czech investors This report is intended for clients of UniCredit Bank AG, UniCredit Bank AG London Branch, UniCredit CAIB AG, UniCredit CAIB Securities UK Ltd. or UniCredit Bank AG Milan Branch in the Czech Republic and may not be used or relied upon by any other person for any purpose. Notice to Italian investors This document is not for distribution to retail clients as defined in article 26, paragraph 1(e) of Regulation n approved by CONSOB on October 29, In the case of a short note, we invite the investors to read the related company report that can be found on UniCredit Research website Notice to Russian investors As far as we are aware, not all of the financial instruments referred to in this analysis have been registered under the federal law of the Russian Federation On the Securities Market dated April 22, 1996, as amended, and are not being offered, sold, delivered or advertised in the Russian Federation. Notice to Turkish investors Investment information, comments and recommendations stated herein are not within the scope of investment advisory activities. Investment advisory services are provided in accordance with a contract of engagement on investment advisory services concluded with brokerage houses, portfolio management companies, non-deposit banks and the clients. Comments and recommendations stated herein rely on the individual opinions of the ones providing these comments and recommendations. These opinions may not suit your financial status, risk and return preferences. For this reason, to make an investment decision by relying solely on the information stated here may not result in consequences that meet your expectations. Notice to Investors in Japan This document does not constitute or form part of any offer for sale or subscription for or solicitation of any offer to buy or subscribe for any securities and neither this document nor any part of it shall form the basis of, or be relied on in connection with or act as an inducement to enter into, any contract or commitment whatsoever. Notice to UK investors This communication is directed only at clients of UniCredit Bank AG, UniCredit Bank AG London Branch, UniCredit CAIB AG, UniCredit CAIB Securities UK Ltd. or UniCredit Bank AG Milan Branch who (i) have professional experience in matters relating to investments or (ii) are persons falling within Article 49(2)(a) to (d) ( high net worth companies, unincorporated associations, etc. ) of the United Kingdom Financial Services and Markets Act 2000 (Financial Promotion) Order 2005 or (iii) to whom it may otherwise lawfully be communicated (all such persons together being referred to as relevant persons ). This communication must not be acted on or relied on by persons who are not relevant persons. Any investment or investment activity to which this communication relates is available only to relevant persons and will be engaged in only with relevant persons. Notice to U.S. investors This report is being furnished to U.S. recipients in reliance on Rule 15a-6 ("Rule 15a-6") under the U.S. Securities Exchange Act of 1934, as amended. Each U.S. recipient of this report represents and agrees, by virtue of its acceptance thereof, that it is such a "major U.S. institutional investor" (as such term is defined in Rule 15a-6) and that it understands the risks involved in executing transactions in such securities. Any U.S. recipient of this report that wishes to discuss or receive additional information regarding any security or issuer mentioned herein, or engage in any transaction to purchase or sell or solicit or offer the purchase or sale of such securities, should contact a registered representative of UniCredit Capital Markets, Inc. ( UCI Capital Markets ). Any transaction by U.S. persons (other than a registered U.S. broker-dealer or bank acting in a broker-dealer capacity) must be effected with or through UCI Capital Markets. The securities referred to in this report may not be registered under the U.S. Securities Act of 1933, as amended, and the issuer of such securities may not be subject to U.S. reporting and/or other requirements. Available information regarding the issuers of such securities may be limited, and such issuers may not be subject to the same auditing and reporting standards as U.S. issuers. The information contained in this report is intended solely for certain "major U.S. institutional investors" and may not be used or relied upon by any other person for any purpose. Such information is provided for informational purposes only and does not constitute a solicitation to buy or an offer to sell any securities under the Securities Act of 1933, as amended, or under any other U.S. federal or state securities laws, rules or regulations. The investment opportunities discussed in this report may be unsuitable for certain investors depending on their specific investment objectives, risk tolerance and financial position. In jurisdictions where UCI Capital Markets is not registered or licensed to trade in securities, commodities or other financial products, transactions may be executed only in accordance with applicable law and legislation, which may vary from jurisdiction to jurisdiction and which may require that a transaction be made in accordance with applicable exemptions from registration or licensing requirements. The information in this publication is based on carefully selected sources believed to be reliable, but UCI Capital Markets does not make any representation with respect to its completeness or accuracy. All opinions expressed herein reflect the author s judgment at the original time of publication, without regard to the date on which you may receive such information, and are subject to change without notice. UCI Capital Markets may have issued other reports that are inconsistent with, and reach different conclusions from, the information presented in this report. These publications reflect the different assumptions, views and analytical methods of the analysts who prepared them. Past performance should not be taken as an indication or guarantee of future performance, and no representation or warranty, express or implied, is provided in relation to future performance. UCI Capital Markets and any company affiliated with it may, with respect to any securities discussed herein: (a) take a long or short position and buy or sell such securities; (b) act as investment and/or commercial bankers for issuers of such securities; (c) act as market makers for such securities; (d) serve on the board of any issuer of such securities; and (e) act as paid consultant or advisor to any issuer. The information contained herein may include forward-looking statements within the meaning of U.S. federal securities laws that are subject to risks and uncertainties. Factors that could cause a company s actual results and financial condition to differ from expectations include, without limitation: political uncertainty, changes in general economic conditions that adversely affect the level of demand for the company s products or services, changes in foreign exchange markets, changes in international and domestic financial markets and in the competitive environment, and other factors relating to the foregoing. All forward-looking statements contained in this report are qualified in their entirety by this cautionary statement This document may not be distributed in Canada or Australia. UniCredit Research page 13

14 UniCredit Research* Thorsten Weinelt, CFA Global Head of Research & Chief Strategist Dr. Ingo Heimig Head of Research Operations Economics & FI/FX Research Marco Annunziata, Ph.D., Chief Economist Economics & Commodity Research Global Economics Dr. Davide Stroppa, Global Economist European Economics Andreas Rees, Chief German Economist Marco Valli, Chief Italian Economist Stefan Bruckbauer, Chief Austrian Economist Tullia Bucco Chiara Corsa Dr. Loredana Federico Alexander Koch, CFA Chiara Silvestre US Economics Dr. Harm Bandholz, CFA Commodity Research Jochen Hitzfeld Nikolaus Keis EEMEA Economics & FI/FX Strategy Cevdet Akcay, Ph.D., Chief Economist, Turkey , Matteo Ferrazzi., Economist, EEMEA , Dmitry Gourov, Economist, EEMEA , Hans Holzhacker, Chief Economist, Kazakhstan , Anna Kopetz, Economist, Baltics , Marcin Mrowiec, Chief Economist, Poland , Vladimir Osakovsky, Ph.D., Head of Strategy and Research, Russia ext.7558, Rozália Pál, Ph.D., Chief Economist, Romania , Kristofor Pavlov, Chief Economist, Bulgaria , kristofor.pavlov@unicreditgroup.bg Goran Šaravanja, Chief Economist, Croatia , goran.saravanja@unicreditgroup.zaba.hr Pavel Sobisek, Chief Economist, Czech Republic , pavel.sobisek@unicreditgroup.cz Gyula Toth, Economist/Strategist, EEMEA , gyula.toth@caib.unicreditgroup.eu Jan Toth, Chief Economist, Slovakia , jan.toth@unicreditgroup.sk Global FI/FX Strategy Michael Rottmann, Head , michael.rottmann1@unicreditgroup.de Dr. Luca Cazzulani, Deputy Head, FI Strategy , luca.cazzulani@unicreditgroup.de Chiara Cremonesi, FI Strategy , chiara.cremonesi@unicreditgroup.eu Dr. Stephan Maier, FX Strategy , stephan.maier@unicreditgroup.eu Giuseppe Maraffino, FI Strategy , giuseppe.maraffino@unicreditgroup.de Armin Mekelburg, FX Strategy , armin.mekelburg@unicreditgroup.de Roberto Mialich, FX Strategy , roberto.mialich@unicreditgroup.de Kornelius Purps, FI Strategy , kornelius.purps@unicreditgroup.de Herbert Stocker, Technical Analysis , herbert.stocker@unicreditgroup.de Publication Address UniCredit Research Corporate & Investment Banking UniCredit Bank AG Arabellastrasse 12 D Munich Tel Fax Bloomberg UCGR Internet * UniCredit Research is the joint research department of UniCredit Bank AG (UniCredit Bank), UniCredit CAIB Group (UniCredit CAIB), UniCredit Securities (UniCredit Securities), UniCredit Menkul Değerler A.Ş. (UniCredit Menkul), UniCredit Bulbank, Zagrebačka banka, UniCredit Bank, Bank Pekao, Yapi Kredi, UniCredit Tiriac Bank and ATFBank. UniCredit Research page 14